Composition of skeletons and support theorems (Q4357550)

Let \(H\) be a real separable Hilbert space, and \(W=\{W(h),\;h\in H\}\) a Gaussian centered family, with \(E[W(h)W(g)]=\langle h,g\rangle_H\), defined on a probability space \((\Omega,\mathcal F,P)\), where \(\mathcal F\) is the \(P\)-completion of \(\sigma\{W\}\). Given a random variable \(G\) with values in a Polish space \((S,d)\), a function \(\Psi: H\rightarrow S\) is a skeleton of \(G\) if: (i) there exists a sequence \(\{\Pi^N\}_N\) of finite-dimensional linear mappings on \(H\), converging strongly to the identity, such that \(\forall \varepsilon>0\), \(\lim_N P\{d(\Psi(W\circ\Pi^N),G)>\varepsilon\}=0\), and (ii) for each \(h\in H\), there exists a sequence of absolutely continuous transformations \(\{T_N^h\}_N\) on \(\Omega\) such that \(\forall \varepsilon>0\), \(\lim_N P\{d(G\circ T^h_N,\Psi(h))>\varepsilon\}=0\). The utility of this concept is that the support of the law of \(G\) can be characterized as the closure in \(S\) of the set \(\{\Psi(h),\;h\in H\}\) [see \textit{A. Millet} and \textit{M. Sanz-Solé}, in: Séminaire de probabilités XXVIII. Lect. Notes Math. 1583, 36-48 (1994; Zbl 0807.60073)]. Suppose \(F: \Omega\times \mathbb{R}^m\rightarrow S\) is measurable, and continuous in the second argument, and \(\xi: \Omega \rightarrow\mathbb{R}^m\) is a random variable. Consider the composition \(G(\omega)=F(\omega,\xi(\omega))\). Given skeletons \(h\mapsto \Phi(h,x)\) for \(F(\cdot,x)\) and \(h\mapsto\varphi(h)\) for \(\xi\), it is shown that \(\Phi(h,\varphi(h))\) is a skeleton for \(G\) under the assumption that \(\Phi(\cdot,x)\) is a uniform skeleton with respect to \(x\), that means, the limits in (i) and (ii) are replaced by \(\lim_N P\{\sup_{|x|\leq K} d(\Psi(W\circ\Pi^N,x),F(x))>\varepsilon\}=0\) and \(\lim_N P\{\sup_{|x|\leq K} d(F(T_N^h,x),\Phi(h,x))>\varepsilon\}=0\) for all \(K\) compact, respectively. An explicit uniform skeleton is shown for a diffusion process starting at \(x\in\mathbb{R}^m\). As an application of the composition result, this allows to deduce a support theorem for anticipating stochastic differential equations when the initial condition is a finite sum of multiple stochastic integrals with continuous kernels. Finally, the existence of a skeleton for a random variable in the second Wiener chaos is proved, without assuming continuity of the kernel. For the entire collection see [Zbl 0870.00040].